Skip to content

Features

Fastkit-Mobility runs the scikit-mobility measure API on a Rust core, on whatever dataframe you already use.

623x median speedup on 16 measures

Familiar API →

Port existing code with an import change.

Any dataframe →

pandas, Polars and other Narwhals backends in; the same backend out.

Low memory

Kernels read your columns in place instead of copying or sorting whole tables.

Validated

Correctness tests on pandas and Polars, plus comparison tests against scikit-mobility. Run them yourself.

Open source

BSD-3 licensed. Read the source, file issues, reproduce every benchmark.

Any dataframe in, same backend out

Data moves from pandas or Polars through Narwhals and Arrow buffers into a Rust kernel, then returns as the same backend type

Switching is one import

# before
import skmob.measures.individual as m
# after
import fastmob as m

Speed at a glance

Bar chart comparing scikit-mobility and fastmob execution time on 4M check-ins for five measures; fastmob is 423x to 2658x faster.

Average of 5 runs on 4 million Brightkite check-ins, pandas backend. Hardware and methodology are on the benchmarks page.

Dig deeper

Benchmarks →

Numbers for every measure family, pandas and Polars.

Optimizations →

How indexed ranges and batched kernels keep memory low.